[
  {
    "benchmark": "longbench",
    "retention": 1.0312,
    "token_savings_pct": 85.3,
    "baseline_accuracy": 0.64,
    "baseline_ci_95": [
      0.5014,
      0.7586
    ],
    "entroly_accuracy": 0.66,
    "entroly_ci_95": [
      0.5215,
      0.7756
    ],
    "baseline_avg_tokens": 12652.3,
    "entroly_avg_tokens": 1860.1,
    "samples": 50,
    "provenance": {
      "model": "gpt-4o-mini",
      "seed": 42,
      "budget": 2000,
      "timestamp_utc": "2026-05-24T01:32:48.398421+00:00",
      "git_sha": "0a54982698ee",
      "elapsed_s": 161.8,
      "audit_jsonl": "longbench_audit.jsonl",
      "audit_record_count": 50,
      "scoring": "bench.accuracy._check_answer (benchmark='longbench')",
      "compressor": "entroly.qccr.select (via bench.accuracy._compress_messages_modal mode='entroly')",
      "dataset_loader": "_load_longbench"
    }
  }
]